6 ms·
Because the alternative methods for handling intra-container communication are even bigger messes. The 1.6 docker solution involved a double nat and relied on
by falcolas 11y ago
Because the alternative methods for handling intra-container communication are even bigger messes.
The 1.6 docker solution involved a double nat and relied on iptables - resulting in some fairly serious bottlenecks and pathalogical edge cases. It also required a third party solution for handling discovery of IPs for services.
Opening up containers to access the host network interfaces breaks the encapsulation promises of containers, and is thus not available to all people. Conceptually, it also creates holes in the idempotent service model, since they have to be aware of port conflicts.
The 1 IP per container model, such as VXLan, Flannel, Kubernetes, Docker 1.7 etc. is one of the more effective methods of countering the problem, at the cost of guzzling IP address space, and requiring a gateway to escape the virtual network tunnel.
- otterley 11y ago> Opening up containers to access the host network interfaces breaks the encapsulation promises of containers Who made that promise? It was never a feature of Linux containers to virtualize the NIC.
- jpgvm 11y agoYou would be surprised the number of people that don't understand containers are a form namespacing + isolation not a form of virtualisation.